Development and Validation of Sandwich ELISA Microarrays with Minimal Assay Interference Rachel M. Gonzalez,† Shannon L. Seurynck-Servoss,† Sheila A. Crowley,† Marty Brown,† Gilbert S. Omenn,§ Daniel F. Hayes,† and Richard C. Zangar*,† Pacific Northwest National Laboratory, 902 Battelle Boulevard, P7-56, Richland, Washington 99354, Breast Oncology Program, University of Michigan Comprehensive Cancer Center, 1500 East Medical Center Drive, Ann Arbor, Michigan 48109, and University of Michigan Medical School, A510 MSRB I, Ann Arbor, Michigan 48109 Received December 6, 2007
Sandwich enzyme-linked immunosorbent assay (ELISA) microarrays are emerging as a strong candidate platform for multiplex biomarker analysis because of the ELISA’s ability to quantitatively measure rare proteins in complex biological fluids. Advantages of this platform are high-throughput potential, assay sensitivity and stringency, and the similarity to the standard ELISA test, which facilitates assay transfer from a research setting to a clinical laboratory. However, a major concern with the multiplexing of ELISAs is maintaining high assay specificity. In this study, we systematically determine the amount of assay interference and noise contributed by individual components of a multiplexed 24-assay system. We find that nonspecific reagent cross-reactivity problems are relatively rare. We did identify the presence of contaminant antigens in a “purified antigen”. We tested the validated ELISA microarray chip using paired serum samples that had been collected from four women at a 6-month interval. This analysis demonstrated that protein levels typically vary much more between individuals than within an individual over time, a result which suggests that longitudinal studies may be useful in controlling for biomarker variability across a population. Overall, this research demonstrates the importance of a stringent screening protocol and the value of optimizing the antibody and antigen concentrations when designing chips for ELISA microarrays. Keywords: antibody • ELISA • microarray • cross-reactivity • longitudinal • serum
Introduction In cancer research, discovering biomarkers that would function as general screening tests in diagnostic or prognostic assay is a primary goal. However, single biomarkers are probably limited in their usefulness due to the heterogeneity of tumors and patient populations. The use of DNA microarray technology has shown that molecular profiling of gene expression patterns can be superior for diagnosis or prognosis when compared to traditional analyses.1,2 For example, gene profiling studies have shown an improved ability to predict the outcome of a treatment protocol in patients with breast cancer.2–5 However, gene array technology is not suited for identifying disease markers in blood, and the use of multiplex protein screens for individualized profiling is currently limited by throughput, lack of sensitivity, or cost (reviewed in refs 6–9 and see refs 10–15). Studies with sandwich enzyme-linked immunosorbent assay (ELISA) microarrays have suggested that this platform has potential as a screening tool for validating candidate biomarkers.16 The ELISA microarrays have several significant advan* Corresponding author. Phone: 1-509-376-8596. Fax: 1-509-376-6767. E-mail:
[email protected]. † Pacific Northwest National Laboratory. † University of Michigan Comprehensive Cancer Center. § University of Michigan Medical School.
2406 Journal of Proteome Research 2008, 7, 2406–2414 Published on Web 04/19/2008
tages over current DNA and protein tests. First, the ELISA microarray is a highly stringent assay with exceptional sensitivity and specificity because it requires two antibodies for detection of each analyte (reviewed in ref 17). Second, the miniature assay design makes it cost-effective for simultaneously screening hundreds of proteins in a high-throughput manner.17 Third, the ELISA microarray can quantitatively measure rare proteins in complex biological fluids. Finally, the similarity to current clinical ELISA protocols will facilitate the transition of promising assays into clinical applications. Sandwich ELISA microarrays are susceptible to measurement error associated with chip design and reagent cross-reactivity. It is because of these factors that nonspecific assay problems are introduced. Therefore, one of the greatest challenges associated with the multiplexing of ELISA tests is the need to optimize the assay components in order to maintain assay specificity. Thus, to achieve quantitative protein measurements in a microarray format requires rigorous attention to chip design and procedural aspects to minimize the risk of bias or erroneous results. In this study, we systematically evaluate reagent specificity for 25 ELISA tests on a single microarray chip. We undertake a variety of screens to define antibody and antigen specificity. We then evaluated serum samples that were collected six months apart which showed large concentration heterogeneity 10.1021/pr700822t CCC: $40.75
2008 American Chemical Society
research articles
Development and Validation of Sandwich ELISA Microarrays a
Table 1. Reagent Sources, Catalog Numbers, and Abbreviated Names analyte
abbr.
capture antibody
source
detection antibodyb
source
purified protein
source
amphiregulin cancer antigen 15-3 cathepsin L CD14 epidermal growth factor heparin-binding epidermal growth factor epidermal growth factor receptor E-selectin basic fibroblast growth factor c-erbB-2 extracellular domain hepatocyte growth factor intercellular adhesion molecule 1 insulin-like growth factor 1 interleukin 1-alpha interleukin 18 matrix metalloprotease 1 matrix metalloprotease 2 matrix metalloprotease 9 platelet-derived growth factor AA prostate specific antigen (hK13) RANTES (CCL5) transforming growth factor alpha tumor necrosis factor alpha urokinase-type plasminogen receptor vascular endothelial growth factor A
AmR CA15-3 CatL CD14 EGF HB-EGF EGFR Esel FGFb HER2 HGF ICAM IGF1
MAB262 M81071022 MAB952 MAB3833 DY236 kit AF-292 AF-231 AF-724 MAB233 MS-229 MAB694 MAB720 MAB291
1 2 1 1 1 1 1 1 1 3 1 1 1
BAF262 M81071021 BAF952 BAF383 DY236 kit BAF259 BAF231 BAF274 BAM233 BAF1129 BAF294 BAF720 BAF291
1 2 1 1 1 1 1 1 1 1 1 1
262-AR 30-AC17 952-CY 383-CD DY236 kit 259-HE 1095-ER ADP1 30-AF17 1129-ER 294-HGN ADP-4-050 291GF
1 2 1 1 1 1 1 1 2 1 1 1 1
IL-1R IL-18 MMP1 MMP2 MMP9 PDGF PSA RANTES TGFR TNFR uPAR VEGF
MAB200 D044-3 AF901 AF902 AF911 MAB221 MAB1344 MAB678 AF-239 MAB610 MAB807 AF-293
1 4 1 1 1 1 1 1 1 1 1 1
BAF200 D045-6 BAF901 BAF902 BAF911 BAF221 BAF1344 BAF278 BAF239 BAF210 BAF207 BAF293
1 4 1 1 1 1 1 1 1 1 1 1
200-LA B003-5 901-MP 902-MP 911-MP 221-AA J62800089 278-RN/CF 293A 210-TA 807-UK 293-VE
1 4 1 1 1 1 5 1 1 1 1 1
a Source code: (1) R&D Systems; Minneapolis, MN, USA. (2) Fitzgerald Industries; Concord, MA, USA. (3) NeoMarkers; Fremont, CA, USA. (4) Medical & Biological Labs; Woburn, MA, USA. (5) BiosPacific; Emeryville, CA, USA. b All detection antibodies are biotinylated by the manufacturer except CA15-3.
between patients but little change between time points. Overall, these screens indicate that, for most assays in a microarray platform, multiplex ELISA analysis can be conducted with minimal assay interference. However, we do find one example each of nonspecific binding between antibodies and reagent contamination. These studies highlight the need for careful evaluation of ELISA reagents prior to use in a multiplex analysis.
Materials and Methods Antibodies and Recombinant Proteins. The components of each assay, that is, the antigen and capture and detection antibodies, were generally purchased from the same source and had previously been demonstrated to work as a sandwich ELISA. The antibodies and antigens used in the ELISA microarrays are listed in Table 1. Our chip contains both monoclonal and polyclonal antibodies developed from multiple animal sources. All detection antibodies were purchased biotinylated except for CA15-3, which was labeled using the Pierce EZ link Sulfo-NHS Biotinylation Kit according to the manufacturer’s protocol. Manufacturing Antibody Chips. ELISA microarray manufacture was undertaken as previously described in detail.18 In brief, we used activated aminosilanated glass slides stamped with a hydrophobic barrier to create 16 wells on each slide (Erie Scientific, Portsmouth, NH). The capture antibodies were suspended in phosphate buffer saline, pH 7.2 (PBS), at concentrations from 0.5 to 1.0 mg/mL. The antibodies were printed on 25 × 75 mm glass slides using a noncontact NanoPlotter NP2 printer (GeSiM, Germany). Approximately 400 picoliters were printed per spot at a pitch of 500 µm. Sixteen identical chips were printed on each slide, with each chip composed of fluorescently labeled nonimmune immunoglobins, which were used as orientation spots, and the individual capture antibod-
ies. The reagents were printed in triplicate or quadruplicate on each chip, with each capture antibody being printed either in a consecutive row or once in four identical quadrants. Proper printing was confirmed using the Red Reflect option on the ScanArray ExpressHT (PerkinElmer, Santa Clara, CA). The slides were blocked in 1% casein in PBS (BioRad, Hercules, CA) except when otherwise noted in the text. Slides were then stored at -20 °C under desiccant and vacuum until use. Preparation of the Antigen and Detection Antibody Mixtures. A mixture containing all the antigens, diluted in 0.1% casein in PBS unless noted otherwise, was prepared using the antigen concentrations listed in Table 2. The antigen mixture was typically prepared without PSA and stored as single-use aliquots at -80 °C. The PSA antigen, which is not stable in a frozen diluted state, was added just prior to use from a stock solution stored at 4 °C. In some cases, to evaluate individual reagents, special antigen mixtures were prepared which contained either a single antigen or the complete mixture minus a single antigen. To generate standard curves, eight point titrations were made using 3- or 4-fold serial dilutions of the standard mixture in 0.1% casein/PBS. Tests performed using only a single concentration of the antigen mix were done at 11% of the maximal concentration listed in Table 2. Similar to the antigens, all detection antibodies were combined in a single solution in 0.1% casein in PBS, unless noted otherwise. To evaluate individual reagents, special mixtures were prepared which contained either a single detection antibody or the complete mixture minus one antibody. Concentrations of the detection antibodies were varied as indicated in the text, but typically the concentration of each antibody was 25 ng/mL. Processing the ELISA Microarray Chips. The protocol for processing the ELISA microarray chips has been previously Journal of Proteome Research • Vol. 7, No. 6, 2008 2407
research articles
Gonzalez et al.
Table 2. Maximal Concentrations of Antigens antigen
maximal pg/mL
predicted mass (KD)a
predicted molarityb
AmR CA15-3 CatL CD14 EGF EGFR E-selectin FGFb HB-EGF Her2 HGF ICAM IGF1 IL-1R IL-18 MMP1 MMP2 MMP9 PDGF PSA RANTES TGFR TNFR uPAR VEGF
1000 1200 units 5000 5000 500 2500 2500 2000 500 5000 1000 7500 2000 500 1000 5000 5000 1000 1000 1000 500 500 500 5000 750
11 109 36 35.8 6 68.6 58.6 16 9.5 96.8 83 50 7.5 18 18 53 71 77 29 26.5 7.8 6 17.5 31 42
9.1 × 10-11 NDc 1.4 × 10-10 1.4 × 10-10 8.3 × 10-11 3.6 × 10-11 4.3 × 10-11 1.2 × 10-10 5.3 × 10-11 5.2 × 10-11 1.2 × 10-11 1.5 × 10-10 2.7 × 10-10 2.8 × 10-11 5.6 × 10-11 9.4 × 10-11 7.0 × 10-11 1.3 × 10-11 3.4 × 10-11 3.8 × 10-11 6.4 × 10-11 8.3 × 10-11 2.9 × 10-11 1.6 × 10-10 1.8 × 10-11
a Predicted mass of expressed protein based on sequence. Many proteins are glycosylated, so the mass for these proteins is an underestimate of the true mass, and the molarity is an overestimate. b Approximate molarity based on mass predicted from sequence. This value will be an overestimation of the true concentration if the antigen is not pure or if it is glycosylated. c ND, not determined.
described in detail.18 Briefly, all incubation steps were performed at room temperature in a closed chamber with saturated humidity and using gentle mixing on an orbital shaker. Washes were performed between each incubation step by submerging the slides in PBS containing 0.05% Tween-20 (PBST) without mixing. The first step was to thaw the slides, dip them in PBS-T, and add 15 µL of sample to each chip. The slides were then incubated overnight. The next step was to add a cocktail of the detection antibodies in 0.1% casein/PBS buffer and incubate for 2 h. The signal was enhanced with the biotinyltyramide amplification system (Perkin-Elmer) as described previously.18,19 Cy3- or Alexa647-conjugated streptavidin was diluted to 1 µg/mL in PBS-T, added to the slide, and incubated for 1 h in the dark. Slides were then quickly rinsed with deionized water and dried. The slides were imaged with ScanArray Express HT laser scanner using either the Cy3 or Alexa 647 settings. ScanArray Express Quantitative software was used to quantify the spot fluorescence intensity from the scanned images. Effect of Serum on Antigen Concentration Measurements. Human blood serum from an anonymous female donor (#M99869, Golden West Biologicals, Temecula, CA) was used. To remove particulate matter, the serum was spun at 15 000g for 20 min at 4 °C, and the supernatant was used for analysis. To prevent saturation of most of ELISAs by antigens present in the serum, serum was diluted to 1% in PBS/0.1% casein. A mixture of all the antigens except CA15-3 and CatL (see Table 1) was added to the diluted serum solution and in the same solution without serum. The concentration of each of the spiked antigens was in the approximate “linear range” of the appro2408
Journal of Proteome Research • Vol. 7, No. 6, 2008
priate standard curve. The diluted “untreated” serum solution was also analyzed without any spiked antigens so that baseline concentration values of each antigen could be determined. Full standard curves were also prepared and analyzed in 0.1% casein/PBS. The samples were processed using the ELISA microarray protocol described above, and the concentrations of the antigens in the spiked and untreated serum samples were calculated using ProMAT. The portion of the measured concentration that was due to the antigen spike was calculated by subtracting the baseline antigen concentration in the untreated serum sample from the value obtained from the spiked serum. The effect of the serum on each measured antigen concentration was then determined by dividing this differential value by the value obtained in the absence of serum. Patients and Serum Collection. Samples for this study were obtained from four women with a history of ductal carcinoma in situ (DCIS) of the breast. These four women were selected from a larger group of 38 women who were enrolled in a prospective study evaluating changes in mammographic density and ductal lavage before and after tamoxifen. At baseline, all subjects had a normal mammogram and clinical breast exam within the past year of study enrollment. In addition, subjects had normal values for their complete blood count and differential (CBC and diff) platelet count, prothrombin time international normalized ratio (INR), aspartate transaminase (AST), alanine transaminase (ALT), total bilirubin, alkaline phosphatase, and creatinine. The average age of the four study participants was 55 years (median ) 55.5, range ) 42-67 years). Three participants were Caucasian, and one was African-American. The average age of menses was 13 years, with half of the participants premenopausal and the other half postmenopausal. The average age of first live birth for the participants was 24 years (range ) 19-29 years). None of the participants had a first degree relative with a history of breast cancer, although half had at least one second degree relative with a breast cancer diagnosis. None of the participants were tested for BRCA1-2. One participant had a one-year history of using hormone replacement therapy, while two participants used oral contraceptive pills for an average of 4.5 years. The first 10 mL blood sample was obtained at baseline (before the start of the tamoxifen treatment) and the second approximately 6 months later. The sample was spun down at 1200 rpm for 20 min, no later than 1 h after the draw. The serum was collected in a plastic transfer pipet and aliquoted into three to five 1.5 mL centrifuge tubes. Samples were stored at -20 °C overnight and then at -80 °C. Prior to microarray analysis, serum samples were thawed and centrifuged at 21 000g for 30 min to remove particulates. Samples were then serially diluted in 0.1% casein in PBS to obtain dilutions ranging from 5- to 4000-fold. Samples were then processed using the ELISA microarrays as described above. For each of the 24 assays, the sample dilution that resulted in signal intensity values that was best characterized by the useful range of that assay’s standard curve was used to calculate sample concentrations for that assay. Data Analysis. Standard curves were generated and graphed using the Protein Microarray Analysis Tool (ProMAT),20,21 which is a free software program we developed specifically for the analysis of ELISA microarray data (http://www.pnl.gov/statistics/ ProMAT/). ProMAT was used to generate standard curves based on a four-parameter logistic model and to calculate the lower limits of detection. The lower limits of detection for the
Development and Validation of Sandwich ELISA Microarrays
research articles
standard curves were calculated as the mean plus three standard deviations of the log-transformed spot intensity for the antigen-free blank.
Results Optimization of the Antigen and Detection Antibody Concentrations. The “four parameters” that are used to fit the standard data with the four-parameter logistics model are the upper and lower asymptotes, the midpoint of the symmetrical “S-shaped” curve, and the slope of the curve at the midpoint. To properly define this curve, the highest concentration for each antigen is selected with the goal of producing a signal that is near saturation but will be in the upper usable range of the standard curve by the third dilution of a 3-fold dilution series, which is equivalent to 11% of the maximal concentration shown in Table 2. Such a dilution series allows for reasonably accurate estimations of all four parameters of the logistic curve. Examples of suitable curves can be seen in Figure 1A. The next step is to optimize the signal-to-noise ratio for each assay by adjusting the concentration of the detection antibodies. This must be done in the multiplex format since the background signal in the presence of multiple reagents will be greater than in a single assay format. For this test, the assay signal was determined for antigens at 11% of the maximal concentration shown in Table 2. This concentration ensures that a strong fluorescent signal is obtained for each assay while maintaining antigen concentrations in the upper range of the usable portion of the standard curve. The background signal or “noise” was determined in chips that were processed without any antigens. The detection antibodies were tested at 100, 50, 25, 12.5, and 6.25 ng/mL. The maximal signal-to-noise ratio for the detection antibody was typically observed at 25 ng/mL or less. The results from this analysis, which are split into two separate graphs for easier interpretation, are shown in Figure 1B and 1C. For subsequent studies, individual detection antibodies were typically used at 25 ng/mL. Representative standard curves that were generated using 25 ng/mL of detection antibody from a single multiplex analysis are shown in Supplemental Figure 1. Potential Interactions between Assays. When using multiplexed ELISAs, there is a potential for the reagents from one assay to interfere with another assay. Examples of an ideal sandwich ELISA and various possibilities for reagent interactions are illustrated in Figure 2. One possibility for assay interference is direct interaction between capture and detection antibodies (Figure 2B). In this case, the signal intensity is essentially independent of antigen concentrations. It is also possible that a detection antibody nonspecifically binds to another targeted antigen or that that two antigens will interact (Figure 2C and D, respectively). In these two cases, the signal is influenced by both the targeted and the interacting antigen levels. An analogous situation is where the detection antibody nonspecifically binds a different antigen (Figure 2E). In this case, however, signal intensity remains proportional to the targeted antigen concentration in both the standards and the biological samples. Therefore, this nonspecific assay interaction typically would not compromise the accuracy or specificity of the assay. In the following sections, we systematically analyze for all possible nonspecific interactions shown in Figure 2 using a series of overlapping tests. Specificity of Individual Components of the ELISA Microarray System. In the first test, individual antigen and detection antibody pairs were assayed on chips that contain all 24 capture
Figure 1. Optimization of detection antibody concentrations. (A) A representative standard curve for AmR that was generated from 3-fold dilution of the purified antigen mix. The curve was fit to a four-parameter logistic curve using ProMAT software. The red arrow indicated the dilution that is 11% of the max concentration. (B and C) Signal-to-noise ratios for each antigen over a range of detection antibody concentrations are shown. Values are shown as the percentage of the maximal signal-to-noise ratio for each assay. See Table 2 for a list of assay abbreviations.
antibodies. If either the antigen or the detection antibody crossreacted with any capture antibodies (e.g., Figure 2B or C, respectively), the cross-reactivity would be readily detected as Journal of Proteome Research • Vol. 7, No. 6, 2008 2409
research articles
Gonzalez et al.
Figure 2. Possible interactions by components of multiplexed ELISAs. (A) The ideal ELISA microarray interaction; (B) direct interactions of the capture antibody (CapAb) and detection antibody (DetAb); (C) nonspecific and specific interactions between a capture antibody and antigens (Ag); (D) nonspecific interactions between two antigens; and (E) nonspecific interactions between a detection antibody and an antigen.
Figure 3. Cross-reactivity of capture antibodies with single pairs of antigen and detection antibody. (A) Graphic representation of all data obtained from individual analysis of the 24 pairs of antigens and detection antibodies using chips that contain all 24 capture antibodies. The arrow indicates the one clear example of assay cross-reactivity detected in this analysis, which was produced by interaction of the FGFb reagents with the CatL detection antibody. (B) The right panel shows that the FGFb detection antibody binds directly to the CatL capture antibody in the presence of no antigens. Compare this signal to the spot image on the left, which is representative of normal background levels observed with the CatL or any the other detection antibody when no antigen is present. See Table 2 for a list of assay abbreviations.
an increased signal in the nonspecific capture antibody. The results of this test are shown in Figure 3A. The only clear evidence of assay cross-reactivity from this test was between the FGFb assay and the CatL capture antibody. To distinguish whether this interaction was due to antigen or detection antibody nonspecificity, we ran a chip with no antigens and probed with only the FGFb detection antibody. Figure 3B shows a strong signal between the FGFb detection antibody and the 2410
Journal of Proteome Research • Vol. 7, No. 6, 2008
CatL capture antibody when no antigens were added. These results suggest that the FGFb detection antibody is directly binding to the CatL capture antibody, as shown in Figure 2B. Evaluation of Reagent Contamination. We evaluated the possibility of nonspecific antigen binding or antigen contamination by screening individual antigens using the complete mixture of detection antibodies. These analyses did not detect any cross-reactivity except in the case of the
Development and Validation of Sandwich ELISA Microarrays
research articles
Figure 4. Contamination of the CA15-3 antigen. Chips were incubated only with the CA15-3 antigen and then processed using either a mixture of all detection antibodies except CA15-3 or a single detection antibody, as indicated beneath each image. The upper left panel shows the layout of the capture antibodies on the chips. The images in this figure are slightly overexposed to better develop the spot background signal. This overexposure allows a better comparison between the contaminant signal and the background. See Table 2 for a list of assay abbreviations.
CA15-3 antigen, which interacted with a variety of different assays (Figure 4). The studies conducted with a single antigen and antibody pair (described above and in Figure 3) suggested that this cross-reactivity is not due to the CA15-3 antigen nonspecifically binding to other capture antibodies (e.g., Figure 2C). We therefore incubated chips with CA15-3 as the only antigen and then added single detection antibodies. These analyses clearly demonstrated that the CA15-3 antigen was contaminated with a number of other antigens included on our chip. Therefore, the inclusion of CA15-3 in the antigen mixture caused the standard curve for the contaminant antigens to inappropriately shift to the left. This error would result in the underestimation of the analyte concentrations in the samples that would be proportional to the true antigen concentrations. The contamination problem with the CA15-3 antigen meant that it was not possible to include this antigen in the multiplex analyses described below. Even so, the CA15-3 assay is still suitable for multiplex analysis of clinical samples if the standard curve is generated using only the CA15-3 antigen. It is also important to note that the maximal antigen concentrations shown in Table 2 were determined without CA15-3 interference and therefore are suitable for use in generating standard curves. Removal of Single Reagents. As a final test of our chip, we removed one reagent at a time for what we call “N-1” screens. These screens measure how the complete combination of all the antigens contributes to the overall signal of an individual assay (Figure 5). If the bulk assay mixture does not interfere with an individual assay, then the removal of either the antigen or detection antibody should decrease the assay signal to the same background level as observed when the whole antigen mixture is removed. The N-1 detection antibody tests will also detect assay interactions associated with nonspecific binding of antigens with capture antibodies or other antigens (Figure 2C and D, respectively). The first
tests were conducted with N-1 antigen mixes, where 23 unique antigen mixtures were prepared, each of which lacked a different antigen. None of these mixtures contained the CA15-3 antigen. These chips were analyzed using all 24 detection antibodies. Similar to the N-1 antigen screen, for the N-1 detection antibody analyses, 23 detection antibody mixes were prepared, each of which was missing a different detection antibody but contained the other 23 antibodies including the CA15-3 detection antibody. The average signal intensity of the assay from three replicate chips was used for these analyses. The results from these analyses are shown in Figure 5. No examples of increased background signal were observed with any of the N-1 tests. IL-1R did have a higher background signal than the other assays. This noise is indicative of a generally more “sticky” capture antibody that nonspecifically binds other proteins, particularly the detection antibodies. Even though the IL-1R ELISA has an unusually high background signal, the lower limit of detection for this assay is typically below 10 pg/mL. Therefore, the higher background observed with this assay should not prevent its use as a sensitive assay in the multiplex ELISA system. Quantitative Screen of Proteins in Human Serum. To determine how a complex biological fluid affects the quantitative characteristics of our assay system, we spiked serum with the same antigen mixture we use for our standard curves. The increase due to the spiked antigens was determined by subtracting out the background concentration levels measured in the unaltered serum. Most of the assays were close to the expected 100% value, with an average value of 101% (Table 3). The coefficients of variance (CV) for all of the assays averaged 8%. The lowest recovery values were observed for bFGF and MMP2, which had measured concentrations in serum of about 75% of expected values. The highest recovery value of 134% was for the uPAR assay, which also had the highest CV value Journal of Proteome Research • Vol. 7, No. 6, 2008 2411
research articles
Gonzalez et al.
Figure 5. Evaluation of assay background signal when systematically removing single assay reagents suggests that the antigen or detection antibody mixtures have minimal effect on background signal for the optimized ELISA microarray. For each assay, the first column “All Ag Mix” shows the signal generated when the complete antigen mix and complete detection antibody mix is applied to the chip. The “N-1 Ag” column shows the signal generated when only the antigen for the listed assay is removed from the antigen mix before screening with the complete detection antibody mix. The “Blank” column shows the result of chips that were incubated with blocking buffer containing no antigens and then incubated with detection antibody mix. Finally, the “N-1 DetAb” column shows the signal generated when the complete antigen mix is incubated on the chip but the detection antibody for the listed assay is removed from the detection antibody mix. See Table 2 for a list of assay abbreviations. Table 3. Quantitative Recovery Rates for Each Analyte in the Presence and Absence of 1% Serum assay
% of predicteda
% CV2
AmR CD14 EGF EGFR Esel FGFb HBEGF Her2 HGF ICAM IGF1 IL18 IL1A MMP1 MMP2 MMP9 PDGF PSA RANTES TGFa TNFa uPAR VEGF Average
109 108 87 99 100 76 126 104 119 85 88 101 98 116 74 81 102 111 82 90 107 134 117 101
8.1
